Fabrication of Fictitious Case Law by Artificial Intelligence and Liability Arising from the Use Thereof

Introduction

The utilization of artificial intelligence has become ubiquitous across all fields, and the realm of law has been profoundly affected by this transformation. It has become commonplace for both citizens and legal practitioners to employ artificial intelligence tools in searching for judicial precedents or legal information. However, the veracity of legal outputs generated by artificial intelligence is highly questionable. First and foremost, the reliability of outputs produced by artificial intelligence through scanning the surface web warrants verification, as it generates responses from unverified, open-access information available on the internet. The fundamental issue, however, pertains to whether these outputs are grounded in authentic legal sources. In this article, I shall examine this core problem and the resulting regimes of liability through a concrete case I personally experienced.

A) AI-Generated Fictitious Case Law

It is my considered opinion that even proprietary, subscription-based legal artificial intelligence software remains far from pinpointing relevant judicial precedents on a specific subject matter from its own repository. As for general-purpose artificial intelligence tools scanning open-access web sources—unlike those specifically tailored for legal professionals—they go beyond mere inadequacy and actively fabricate fictitious case law. The public news regarding an investigation launched against an attorney for invoking artificial intelligence-generated fake precedents is well known to all. When I requested an artificial intelligence tool to disclose the source of a cited precedent, the response provided was as follows:

The fact that artificial intelligence software, whose core mechanism ought to be limited to scanning the surface web to deliver pre-existing content to the user, spontaneously engages in fabricating judicial precedents constitutes a major problem. This phenomenon is by no means novel, as similar instances have already been adjudicated in the United States. A notable example is MATA v. Avianca Inc.[1] In said case, ChatGPT produced fictitious precedents, which were subsequently relied upon by attorneys within judicial proceedings; consequently, the presiding judge imposed sanctions on the attorneys for breaching their duty of professional due diligence. Attorneys were not the sole professionals to fall into this trap; prosecutors in the United States have likewise been subjected to administrative and disciplinary investigations owing to artificial intelligence-generated erroneous legal information.[2]

B) Civil and Criminal Liability of AI Software Operators for Fictitious Case Law

1) Legislative Void Regarding Criminal Liability

The root cause of case law fabrication lies within the underlying algorithms of artificial intelligence tools. In our domestic jurisdiction, however, there exists no statutory provision enacting prohibitions such as "artificial intelligence shall not generate fictitious information in response to requests for factual data," nor are there statutory sanctions attached thereto. In point of fact, our legal framework lacks specific legislation governing artificial intelligence. Consequently, normative frameworks defining the operational standards, testing protocols, and related matters of artificial intelligence software have not been established. Naturally, in the absence of such normative benchmarks, criminal or administrative liability for statutory infractions cannot be imputed.

In 2024, the Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law[3] was opened for signature.[4] Article 8 of the Convention imposes an affirmative duty of oversight and supervision upon state parties regarding artificial intelligence technologies. Furthermore, Articles 12, 13, and 16 mandate the establishment of requisite standards and testing mechanisms to ensure the reliability of artificial intelligence systems and their outputs. Although a draft bill introduced in 2024[5] and a well-reasoned legislative proposal in 2025[6] aiming to insert a corresponding provision into the Cybersecurity Law existed in Türkiye, neither proposal achieved statutory enactment.

In the absence of bespoke legislation, any potential offenses must be evaluated under pre-existing statutory provisions pursuant to the principle of legality. Nevertheless, establishing the requisite intent, whether by act or omission, on the part of those responsible for artificial intelligence algorithms with respect to existing criminal offenses presents a formidable evidentiary hurdle. Furthermore, liability disclaimers, non-advisory notices, and accuracy warnings embedded within terms of service generally operate as formidable bars to establishing legal liability. Regarding potential harm resulting from the intentional omission of mandatory testing, the prevailing normative uncertainty surrounding testing protocols is bound to create significant impediments in attributing legal responsibility.

2) Civil Liability

a) Judicial Precedents

Globally, the question of civil liability arising from artificial intelligence-generated fictitious case law remains highly controversial and continues to be actively litigated. For instance, OpenAI has been named as a defendant in claims arising from fictitious precedents.[7] A regional court in Germany adjudicated that Google bears civil liability for erroneous statements generated by its artificial intelligence features.[8] Perhaps the most illustrative example is a suit for damages initiated in the United States, where ChatGPT provided a user with a series of legal recommendations comprised of fictitious precedents, explicitly advising the user to discharge their counsel and proceed pro se. Relying upon this erroneous guidance instead of professional counsel, the user suffered substantial financial detriment amounting to hundreds of thousands of dollars upon losing the underlying lawsuit.[9]

I anticipate that we shall shortly witness lawsuits in Türkiye resulting from adverse judgments attributable to artificial intelligence-generated fictitious case law. Indeed, we shall likely observe actions brought against both attorneys and artificial intelligence operating entities. Commenting on the aforementioned ChatGPT litigation, Kahana notes in a Stanford Law publication that "Nippon Life v. OpenAI looks set to force courts and regulators to define a safe harbor for AI-based legal applications."[10] I share the view that regulatory frameworks of this nature will inevitably be enacted in Türkiye in the near future.

b) Liability Under Turkish Law

Artificial intelligence developers customarily incorporate comprehensive disclaimers of liability regarding their generated outputs. Under Turkish law, Article 115 of the Turkish Code of Obligations establishes gross negligence as an exception to contractual exculpatory clauses. Consequently, any general terms of business or contractual terms purporting to exculpate a party for gross negligence are null and void ab initio. Indeed, general terms of service published by entities such as Google incorporate explicit admissions of liability for gross negligence.

Whether deploying an algorithm that permits artificial intelligence to fabricate fictitious precedents and present them as authentic constitutes gross negligence remains open to debate. As noted supra, the precise standard of care required in this field is not normatively defined. While criteria such as the model agent standard might be utilized to assess whether existing testing and auditing procedures conform to prevailing industry standards, the core dilemma remains that most artificial intelligence tools continue to yield hallucinated outputs notwithstanding rigorous testing.

In conclusion, if the failure to implement—or the inability to implement—preventative measures against the fabrication of case law is deemed to constitute gross negligence, civil liability for damages may attach to the natural or legal person operating the artificial intelligence system. However, the absence of codified standards, coupled with demonstrable efforts in testing and system auditing, presents significant legal hurdles to establishing civil liability.

C) Criminal Liability of Attorneys Utilizing Fictitious Case Law

1) The Offense of Forgery of Official Documents

It is clear that attorneys do not unequivocally qualify as public officials within the contemplation of the Turkish Penal Code. The landmark decision of the General Assembly of Criminal Chambers of the Court of Cassation (E. 2021/43, K. 2021/287) adopts a strict interpretation anchored in the principle of legality, establishing criteria such as the exercise of public authority and holding that an attorney does not inherently assume the status of a perpetrator in the crime of embezzlement. Conversely, jurisprudence exists wherein the Court of Cassation has characterized attorneys as public officials in the context of official document forgery, classifying documents prepared in their professional capacity as official instruments, thereby affirming convictions for forgery of official documents (9th Criminal Chamber, E. 2020/731, K. 2021/2802).

Should the construction of the 9th Criminal Chamber be adopted, an attorney who incorporates artificial intelligence-generated fictitious case law into a court motion could be deemed to possess the status of a perpetrator, with the document falling within the scope of intellectual forgery. The crucial issue to be examined is where the attorney acts with eventual intent, reasoning: "I am unaware whether the precedent is authentic and I do not intend to submit a fake authority, but I shall submit it without verification regardless of the outcome." This scenario represents the most probable occurrence. However, this raises dogmatic disputes as to whether intellectual forgery admits of eventual intent. If direct intent or a specific fraudulent motive is required, an attorney who merely omits to verify the authenticity of artificial intelligence-generated case law would not incur criminal liability under this head. Given the divergence of academic opinion on this point, I shall not elaborate further.

In my view, the definition of public official under Article 6 of the Turkish Penal Code fails to conform to Articles 13 and 38 of the Constitution and warrants constitutional invalidation, having been drafted via an overly synthetic methodology. Although not explicitly codified in the Turkish Penal Code, Article 111-4 of the French Penal Code enshrines the foundational maxim: "La loi pénale est d'interprétation stricte". This principle mandates that criminal provisions be construed narrowly. Therefore, notwithstanding its conceptual friction with the principle of legality, Article 6 of the Turkish Penal Code must be strictly construed. Under a strict interpretation, an attorney—who lacks civil servant status and is not deemed a public official under administrative law—should not be classified as a public official merely by virtue of performing a judicial function.

Furthermore, court pleadings submitted by attorneys fall outside the protective scope of official forgery provisions in the context of intellectual forgery. An attorney's motion does not involve the exercise of public authority, nor does it implicate public trust in the authenticity of instruments as a protected legal interest. Counsel does not act as an infallible arbiter of truth in judicial proceedings. Adjudications are not rendered ex parte on the basis of assertions in pleadings; rather, judgments follow adversarial scrutiny wherein the court independently verifies the factual and legal veracity of the submissions. Consequently, counsel's submission of a non-existent precedent within a motion does not, in my submission, satisfy the statutory elements of official forgery.

2) Misconduct in Public Office

Notwithstanding jurisprudence contending that the relevant offense was impliedly repealed alongside Article 5 of the Turkish Penal Code,[11] Article 62 of the Attorneys' Law currently in force explicitly provides: "An attorney who abuses the duties and powers conferred upon them pursuant to this Law and other statutes in their capacity as an attorney... shall be punished in accordance with Article 257 of the Turkish Penal Code." Settled jurisprudence of the Court of Cassation directly categorizes attorneys as potential perpetrators of misconduct in office under Article 257 of the Turkish Penal Code.[12] Thus, it must be acknowledged that attorneys may qualify as perpetrators of this offense.

In my assessment, misconduct in public office may be committed with eventual intent, as the statutory definition contains no specific qualification necessitating direct intent.[13] Accordingly, an attorney who deploys fictitious case law without verification under a spirit of indifference may incur criminal liability for misconduct in office.[14]

However, statutory conviction requires proof of actual detriment or prejudice. Where no prejudice or harm flows from the submission of the motion, criminal liability will not attach. Conversely, where an adverse judgment is rendered in reliance upon fictitious precedent, actionable prejudice and harm will undoubtedly be established.

D) Statutory Limitation and Regulatory Oversight of AI Deployment in Legal and Medical Domains

In domains where erroneous information carries grave repercussions, such as law and medicine, the deployment of general-purpose software should be statutorily restricted. Legitimacy should be reserved exclusively for domain-specific artificial intelligence software operating under comprehensive liability regimes and specialized regulatory frameworks. The rationales for this necessity are manifold. Just as the unauthorized practice of law and medicine constitute criminal offenses designed to safeguard citizens' rights through qualified professional service, analogous regulatory safeguards must be extended to artificial intelligence services.

Laypersons, and indeed legal professionals as evidenced by the foregoing instances, are susceptible to material error induced by artificial intelligence hallucinations. Artificial intelligence fosters the misconception among the public that accurate legal results are accessible without professional intervention, while inducing in professionals the illusion of achieving work products without necessary labor. However, the utilization of such tools by individuals lacking sufficient capacity for critical self-audit can inflict severe detriment upon themselves and society at large. Medicine directly governs human health, whereas law affects the full spectrum of an individual's legal, financial, and personal standing.

In conclusion, the normative frameworks governing the duties of care and testing protocols incumbent upon artificial intelligence developers and operators—both during development and upon commercial release to end-users—represent an urgent regulatory frontier. While a general duty of professional care exists under general principles of law, it remains evident that existing tools can effortlessly fabricate fictitious case law. The inadequacy of the present legal framework is undeniable. The enactment of specific statutory standards and proportionate statutory sanctions for their breach is imperative.

Regarding penal and administrative sanctions, the principles of culpability and statutory legality demand that the precise boundaries of mandatory testing and compliance auditing be statutorily defined prior to attaching liability to intentional omissions. Similarly, as civil liability for gross versus ordinary negligence hinges upon these normative benchmarks, the legislative codification of these parameters remains an indispensable prerequisite.

Atty. Alp Öztekin, LL.M.

alpozztekin@gmail.com

[1] https://caselaw.findlaw.com/court/us-dis-crt-sd-new-yor/2335142.html

[2] https://calmatters.org/economy/technology/2026/08/ai-prosecutor-errors/

[3] https://rm.coe.int/1680afae3c